A failure detection methodology using new features of acoustic images of a fan matrix

In previous studies of the authors, the acoustic images of a matrix of fans were used to detect operation failures on some fans. Promising results showed that maxima values and positions (azimuth and elevation) parameters of acoustic images of the fan matrix could be used to detect if one of the fans of the matrix is not working properly. The acoustic images were, obtained at different frequencies using a planar array of MEMS microphones. This developed methodology, based on a machine learning SVM algorithm, did not work properly if the matrix has more than one faulty fan. This work analyses the convenience of including other geometrical parameters or features of the acoustic images in the detection methodology. The new information that has been test to be used in the methodology is the centroid of the acoustic images and the energy included in the acoustic images on the vicinity of the real position of the fans in the matrix.